Array indexing with list of strings.

[franciscomalveiro]

Hello! First of all thanks for your work on the development of this project. I have been trying to use your code to generate explanations for an LSTM model.

I have been following the notebook you provide and adapting it to my use case.

To generate global feature explanations, you make use of the function

timeshap/src/timeshap/explainer/feature_level.py

Lines 204 to 215 in b07dd7c

	def feat_explain_all(f: Callable,
	data: Union[List[np.ndarray], pd.DataFrame, np.array],
	feat_dict: dict,
	pruning_data: pd.DataFrame,
	baseline: Union[pd.DataFrame, np.array] = None,
	model_features: List[Union[int, str]] = None,
	schema: List[str] = None,
	entity_col: Union[int, str] = None,
	time_col: Union[int, str] = None,
	append_to_files: bool = False,
	verbose: bool = False,
	) -> pd.DataFrame:

I have been calling with with the arguments suited to my use case, specifically providing it the model_features
argument (a list of strings)

timeshap/src/timeshap/explainer/feature_level.py

Lines 245 to 246 in b07dd7c

	model_features: List[str]
	Features to be used by the model. Requires same order as training dataset

and it generates the following error:

line 327, in feat_explain_all
    sequence = sequence[:, :, model_features]
IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices

This is the line of code:

timeshap/src/timeshap/explainer/feature_level.py

Lines 326 to 327 in b07dd7c

	if model_features:
	sequence = sequence[:, :, model_features]

I have noticed that you create indices above and never use the variable model_features_index

timeshap/src/timeshap/explainer/feature_level.py

Line 317 in b07dd7c

model_features_index, entity_col_index, time_col_index = convert_to_indexes(model_features, schema, entity_col, time_col)

It makes sense that the array should be accessed with these indexes, rather than with a list of strings, is that it?

[JoaoPBSousa]

Hello @franciscomalveiro,

Thank you for noting this issue. You are indeed correct. We should use the model_features_index variable instead of model_features.

We already pushed a new version of TimeSHAP with this fixed.